The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_schematic.py invokes subprocess.run to execute a secondary script (scripts/generate_schematic_ai.py) within the skill's directory. This is used for its stated purpose of generating scientific diagrams but represents a powerful system-level capability.
[EXTERNAL_DOWNLOADS]: The skill performs network operations using the requests library to communicate with the openrouter.ai API. This is a functional requirement for its AI-powered schematic generation and review features.
[DATA_EXFILTRATION]: User-provided diagram descriptions and generated image data are transmitted to an external service (openrouter.ai). While consistent with the skill's purpose, this involves the transfer of research descriptions to a third-party platform.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection in scripts/build_decision_tree.py and scripts/create_cohort_tables.py. These scripts ingest untrusted clinical data and text descriptions and interpolate them directly into LaTeX templates without sanitization or boundary markers. Malicious input containing LaTeX commands could be executed during the document compilation process.
[PROMPT_INJECTION]: There is an inconsistency between the author metadata in SKILL.md (K-Dense Inc.) and the provided context author (tassiovale). Additionally, the code contains references to non-existent model versions (e.g., gemini-3.1) and unconventional branding, which may indicate deceptive or poor-quality metadata.

clinical-decision-support